🔐 Digital Payment Fraud Detection & Investigation System

📌 Overview

This project focuses on building an end-to-end fraud detection and investigation system for digital payment transactions. The goal is to identify suspicious activities, detect fraudulent transactions, and generate actionable risk insights that can support real-world fraud investigation teams in fintech environments.

The system is designed to reflect real industry workflows, including fraud pattern analysis, extreme class imbalance handling, risk modeling, and business-focused evaluation.

---

🎯 Business Objective

• Detect fraudulent transactions from digital payment data
• Identify high-risk transactions using data-driven methods
• Support fraud investigators with a risk-scoring framework
• Handle highly imbalanced real-world fraud data
• Translate model results into business insights and recommendations

---

📊 Dataset

The project uses the Credit Card Fraud Detection dataset (European cardholders, September 2013).

• Total transactions: 284,807
• Fraudulent transactions: 492 (0.172%)
• Features: PCA-transformed components (V1–V28), Time, Amount
• Target: Class (0 = normal, 1 = fraud)

Due to confidentiality, original feature meanings are hidden. This makes the dataset ideal for focusing on pattern detection, anomaly identification, and fraud risk modeling.

Source

OpenML: https://www.openml.org/d/1597

Citation

Andrea Dal Pozzolo, Olivier Caelen, Reid A. Johnson and Gianluca Bontempi. Calibrating Probability with Undersampling for Unbalanced Classification. IEEE CIDM, 2015.

---

🛠️ Tech Stack

• Python
• Pandas, NumPy
• Scikit-learn
• Imbalanced-learn (SMOTE)
• Matplotlib, Seaborn
• Google Colab / Jupyter Notebook

---

🧱 Project Workflow

1. Business understanding & problem framing
2. Data cleaning and preparation
3. Fraud investigation-style EDA
4. Feature engineering for fraud behavior
5. Handling extreme class imbalance
6. Fraud detection model development
7. Model evaluation using fraud-focused metrics
8. Transaction risk scoring system
9. Investigation insights generation
10. Business impact analysis & recommendations

---

🔎 Key Highlights

• Investigation-driven exploratory data analysis
• Extreme class imbalance handling using SMOTE
• Behavioral fraud feature preparation
• Logistic Regression & Random Forest models
• Recall and ROC-AUC focused evaluation
• Risk scoring framework (0–100)
• High-risk transaction identification
• Business-oriented fraud insights

---

📈 Results Summary

• Successfully identified fraudulent transactions from highly imbalanced data
• Achieved strong fraud recall while controlling false alerts
• Built a scalable risk-scoring system for investigation workflows
• Generated actionable insights such as high-risk transactions and potential fraud exposure

(Exact metrics can be found in the notebook.)

---

📁 Repository Structure

digital-payment-fraud-detection/
│
├── data/
│   └── creditcard_csv.csv
│
├── notebooks/
│   └── digital_payment_fraud_detection.ipynb
│
├── sql/
│   └── fraud_queries.sql
│
├── README.md
└── requirements.txt

---

🚀 How to Run

1. Clone the repository
2. Install dependencies

pip install -r requirements.txt

3. Open the notebook

jupyter notebook

4. Run all cells in order

---

💼 Business Impact

This system demonstrates how data-driven fraud detection can:

• Improve early fraud identification
• Reduce financial losses
• Optimize investigator workload
• Enable risk-based transaction monitoring
• Support proactive fraud prevention strategies

---

👤 Author

Saurabh Raj Varma Aspiring Data Scientist | Fraud & Risk Analytics

---